Publications

This page lists selected publications, preprints, and ongoing research manuscripts. My work primarily focuses on medical imaging, computer vision, explainability, and applied AI systems.

Selected Publications

Weakly Supervised Pneumonia Localization from Chest X-Rays Using Deep Neural Network and Grad-CAM Explanations

Kiran Shahi
Preprint / Research Manuscript

This work investigates weakly supervised localization of pneumonia from chest X-rays using classification models and Grad-CAM-based visual explanations. The study examines whether disease-relevant regions can be identified using image-level labels alone, and discusses the role of explanation quality in weakly supervised medical imaging pipelines.

Uncertainty-Aware Lung Segmentation for Chest X-Rays Using nnU-Net v2 Ensemble Analysis

Kiran Shahi
Ongoing Research

This project studies the reliability of automated lung segmentation using ensemble-based uncertainty estimation. It analyzes predictive entropy, ensemble variance, and the relationship between segmentation confidence and error regions under both in-distribution and out-of-distribution evaluation.

Temporal Matting with Explainable Confidence Maps

Kiran Shahi
Proposed Research Direction

This line of work explores temporal video matting with a focus on confidence-aware predictions, temporal consistency, and interpretable uncertainty visualization for human-centered editing systems.

Research Themes

My publication roadmap currently centers on the following themes:

• Weakly supervised medical image analysis
• Explainable AI in healthcare
• Uncertainty-aware segmentation
• Temporal vision and video matting

Notes

A more complete and regularly updated publication list will also be maintained through:

• Google Scholar
• ORCID
• GitHub repositories
• Preprint servers